Search

Effectively controlling systems governed by Partial Differential Equations (PDEs) is crucial in several fields of Applied Sciences and Engineering. These systems usually yield significant challenges to conventional control schemes due to their nonlinear dynamics, partial observability, high-dimensionality once discretized, distributed nature, and the requirement for low-latency feedback control. Reinforcement Learning (RL), particularly Deep RL (DRL), has recently emerged as a promising control paradigm for such systems, demonstrating exceptional capabilities in managing high-dimensional, nonlinear dynamics. However, DRL faces challenges including sample inefficiency, robustness issues, and an overall lack of interpretability. To address these issues, we propose a data-efficient, interpretable, and scalable Dyna-style Model-Based RL framework for PDE control, combining the Sparse Identification of Nonlinear Dynamics with Control (SINDy-C) algorithm and an autoencoder (AE) framework for the sake of dimensionality reduction of PDE states and actions. This novel approach enables fast rollouts, reducing the need for extensive environment interactions, and provides an interpretable latent space representation of the PDE forward dynamics. We validate our method on two PDE problems describing fluid flows - namely, the 1D Burgers equation and 2D Navier-Stokes equations - comparing it against a model-free baseline, and carrying out an extensive analysis of the learned dynamics.

We investigate the momentum fluxes between a turbulent air boundary layer and a growing-breaking wave field by solving the air-water two-phase Navier-Stokes equations through direct numerical simulations (DNS). A fully-developed turbulent airflow drives the growth of a narrowbanded wave field, whose amplitude increases until reaching breaking conditions. The breaking events result in a loss of wave energy, transferred to the water column, followed by renewed growth under wind forcing. We revisit the momentum flux analysis in a high-wind speed regime, characterized by the ratio of the friction velocity to wave speed $u_\ast/c$ in the range $[0.3-0.9]$, through the lens of growing-breaking cycles. The total momentum flux across the interface is dominated by pressure, which increases with $u_\ast/c$ during growth and reduces sharply during breaking. Drag reduction during breaking is linked to airflow separation, a sudden acceleration of the flow, an upward shift of the mean streamwise velocity profile, and a reduction in Reynolds shear stress. We characterize the reduction of pressure stress and flow acceleration through an aerodynamic drag coefficient by splitting the analysis between growing and breaking stages, treating them as separate sub-processes. While drag increases with $u_\ast/c$ during growth, it drops during breaking. Averaging over both stages leads to a saturation of the drag coefficient at high $u_\ast/c$, comparable to what is observed at high wind speeds in laboratory and field conditions. Our analysis suggests this saturation is controlled by breaking dynamics.

Conformal prediction (CP) is a distribution-free framework for achieving probabilistic guarantees on black-box models. CP is generally applied to a model post-training. Recent research efforts, on the other hand, have focused on optimizing CP efficiency during training. We formalize this concept as the problem of conformal risk minimization (CRM). In this direction, conformal training (ConfTr) by Stutz et al.(2022) is a technique that seeks to minimize the expected prediction set size of a model by simulating CP in-between training updates. Despite its potential, we identify a strong source of sample inefficiency in ConfTr that leads to overly noisy estimated gradients, introducing training instability and limiting practical use. To address this challenge, we propose variance-reduced conformal training (VR-ConfTr), a CRM method that incorporates a variance reduction technique in the gradient estimation of the ConfTr objective function. Through extensive experiments on various benchmark datasets, we demonstrate that VR-ConfTr consistently achieves faster convergence and smaller prediction sets compared to baselines.

Dry synthesis is a highly versatile method for the fabrication of nanoporous metal films, since it enables easy and reproducible deposition of single or multi-layer(s) of nanostructured materials that can find intriguing applications in plasmonics, photochemistry and photocatalysis, to name a few. Here, we extend the use of this methodology to the preparation of copper nanoislands that represent an affordable and versatile example of disordered plasmonic substrate. We perform detailed characterizations of the system using several techniques such as spectroscopic ellipsometry, cathodoluminescence, electron energy loss spectroscopy, ultrafast pump-probe spectroscopy and second-harmonic generation with the aim to investigate the optical properties of these systems in an unprecedented systematic way. Our study represents the starting point for future applications of this new disordered plasmonic system ranging from sensing to photochemistry and photocatalysis.

Conventional histopathology has long been essential for disease diagnosis, relying on visual inspection of tissue sections. Immunohistochemistry aids in detecting specific biomarkers but is limited by its single-marker approach, restricting its ability to capture the full tissue environment. The advent of multiplexed imaging technologies, like multiplexed immunofluorescence and spatial transcriptomics, allows for simultaneous visualization of multiple biomarkers in a single section, enhancing morphological data with molecular and spatial information. This provides a more comprehensive view of the tissue microenvironment, cellular interactions, and disease mechanisms - crucial for understanding disease progression, prognosis, and treatment response. However, the extensive data from multiplexed imaging necessitates sophisticated computational methods for preprocessing, segmentation, feature extraction, and spatial analysis. These tools are vital for managing large, multidimensional datasets, converting raw imaging data into actionable insights. By automating labor-intensive tasks and enhancing reproducibility and accuracy, computational tools are pivotal in diagnostics and research. This review explores the current landscape of multiplexed imaging in pathology, detailing workflows and key technologies like PathML, an AI-powered platform that streamlines image analysis, making complex dataset interpretation accessible for clinical and research settings., Comment: 54 pages (39 manuscript + 14 supplementary), 3 figures (figure 1, 2 and supplementary figure 1), 6 Tables (Table 1, 2, 3 and supplementary table 1,2,3)

Modern large language models use a fixed tokenizer to effectively compress text drawn from a source domain. However, applying the same tokenizer to a new target domain often leads to inferior compression, more costly inference, and reduced semantic alignment. To address this deficiency, we introduce Sparse Sinkhorn Token Translation (S2T2). S2T2 trains a tailored tokenizer for the target domain and learns to translate between target and source tokens, enabling more effective reuse of the pre-trained next-source-token predictor. In our experiments with finetuned English language models, S2T2 improves both the perplexity and the compression of out-of-domain protein sequences, outperforming direct finetuning with either the source or target tokenizer. In addition, we find that token translations learned for smaller, less expensive models can be directly transferred to larger, more powerful models to reap the benefits of S2T2 at lower cost.

Recent work has documented striking heterogeneity in the performance of state-of-the-art vision language models (VLMs), including both multimodal language models and text-to-image models. These models are able to describe and generate a diverse array of complex, naturalistic images, yet they exhibit surprising failures on basic multi-object reasoning tasks -- such as counting, localization, and simple forms of visual analogy -- that humans perform with near perfect accuracy. To better understand this puzzling pattern of successes and failures, we turn to theoretical accounts of the binding problem in cognitive science and neuroscience, a fundamental problem that arises when a shared set of representational resources must be used to represent distinct entities (e.g., to represent multiple objects in an image), necessitating the use of serial processing to avoid interference. We find that many of the puzzling failures of state-of-the-art VLMs can be explained as arising due to the binding problem, and that these failure modes are strikingly similar to the limitations exhibited by rapid, feedforward processing in the human brain.

The irrelevant composite operator $T\bar{T}$, constructed from components of the stress-energy tensor, exhibits unique properties in two-dimensional quantum field theories and represents a distinctive form of integrable deformation. Significant progress has been made in understanding the bulk aspects of the theory, including its interpretation in terms of coordinate transformations and its connection to topological gravity models. However, the behavior of $T\bar{T}$-deformed theories in the presence of boundaries and defects remains largely unexplored. In this note, we review analytical results obtained through various techniques. Specifically, we study the $T\bar{T}$-deformed exact g-function within the framework of the Thermodynamic Bethe Ansatz and show that the results coincide with those obtained by solving the corresponding Burgers-type flow equation. Finally, we highlight some potentially significant open problems., Comment: 17 pages, 5 figures

Denoising Diffusion Probabilistic Models (DDPMs) are powerful generative deep learning models that have been very successful at image generation, and, very recently, in path planning and control. In this paper, we investigate how to leverage the generalization and conditional-sampling capabilities of DDPMs to generate complex paths for a robotic end effector. We show that training a DDPM with synthetical and low-quality demonstrations is sufficient for generating nontrivial paths reaching arbitrary targets and avoiding obstacles. Additionally, we investigate different strategies for conditional sampling combining classifier-free and classifier-guided approaches. Eventually, we deploy the DDPM in a receding-horizon control scheme to enhance its planning capabilities. The Denoising Diffusion Planner is experimentally validated through various experiments on a Franka Emika Panda robot.

Bosonic bunching is a term used to describe the well-known tendency of bosons to bunch together, and which differentiates their behaviour from that of fermions or classical particles. However, in some situations perfectly indistinguishable bosons may counter-intuitively bunch less than classical, distinguishable particles. Here we report two such counter-intuitive multiphoton bunching effects observed with three photons in a three-mode balanced photonic Fourier interferometer. In this setting, we show indistinguishable photons actually minimize the probability of bunching. We also show that any non-trivial value of the three-photon collective photonic phase leads to a decreased probability of all photons ending up in the same mode, even as we increase pairwise indistinguishability. Our experiments feature engineering of partial indistinguishability scenarios using both the time and the polarization photonic degrees of freedom, and a polarization-transparent 8-mode tunable interferometer with a quantum-dot source of single photons. Besides the foundational understanding, the observation of these counter-intuitive phenomena open news perspective in devising more efficient ways of routing photons for advantage in metrology and quantum computation., Comment: 10+8 pages, 5+1 figures

Current approaches using sequential networks have shown promise in estimating field variables for dynamical systems, but they are often limited by high rollout errors. The unresolved issue of rollout error accumulation results in unreliable estimations as the network predicts further into the future, with each step's error compounding and leading to an increase in inaccuracy. Here, we introduce the State-Exchange Attention (SEA) module, a novel transformer-based module enabling information exchange between encoded fields through multi-head cross-attention. The cross-field multidirectional information exchange design enables all state variables in the system to exchange information with one another, capturing physical relationships and symmetries between fields. Additionally, we introduce an efficient ViT-like mesh autoencoder to generate spatially coherent mesh embeddings for a large number of meshing cells. The SEA integrated transformer demonstrates the state-of-the-art rollout error compared to other competitive baselines. Specifically, we outperform PbGMR-GMUS Transformer-RealNVP and GMR-GMUS Transformer, with a reduction in error of 88% and 91%, respectively. Furthermore, we demonstrate that the SEA module alone can reduce errors by 97% for state variables that are highly dependent on other states of the system. The repository for this work is available at: https://github.com/ParsaEsmati/SEA, Comment: Accepted in 38th Conference on Neural Information Processing Systems (NeurIPS 2024)

Quantum networks (QNs) promise to enhance the performance of various quantum technologies in the near future by distributing entangled states over long distances. The first step towards this is to develop novel entanglement measures that are both informative and computationally tractable at large scales. While numerous such entanglement measures exist for discrete-variable (DV) systems, a comprehensive exploration for experimentally preferred continuous-variable (CV) systems is lacking. Here, we introduce a class of CV entanglement measures, among which we identify a non-convex entanglement monotone -- the ratio negativity, which possesses a simple, scalable form for characterizing the optimal entanglement swapping of pure Gaussian states. This characterization opens avenues for leveraging statistical physics tools to analyze swapping-protocol-based CV QNs.

Deep state-space models (Deep SSMs) have shown capabilities for in-context learning on autoregressive tasks, similar to transformers. However, the architectural requirements and mechanisms enabling this in recurrent networks remain unclear. This study demonstrates that state-space model architectures can perform gradient-based learning and use it for in-context learning. We prove that a single structured state-space model layer, augmented with local self-attention, can reproduce the outputs of an implicit linear model with least squares loss after one step of gradient descent. Our key insight is that the diagonal linear recurrent layer can act as a gradient accumulator, which can be `applied' to the parameters of the implicit regression model. We validate our construction by training randomly initialized augmented SSMs on simple linear regression tasks. The empirically optimized parameters match the theoretical ones, obtained analytically from the implicit model construction. Extensions to multi-step linear and non-linear regression yield consistent results. The constructed SSM encompasses features of modern deep state-space models, with the potential for scalable training and effectiveness even in general tasks. The theoretical construction elucidates the role of local self-attention and multiplicative interactions in recurrent architectures as the key ingredients for enabling the expressive power typical of foundation models., Comment: 16 pages, 5 figures

A Bernoulli factory is a randomness manipulation routine that takes as input a Bernoulli random variable, outputting another Bernoulli variable whose bias is a function of the input bias. Recently proposed quantum-to-quantum Bernoulli factory schemes encode both input and output variables in qubit amplitudes. This primitive could be used as a sub-routine for more complex quantum algorithms involving Bayesian inference and Monte Carlo methods. Here, we report an experimental implementation of a polarization-encoded photonic quantum-to-quantum Bernoulli factory. We present and test three interferometric set-ups implementing the basic operations of an algebraic field (inversion, multiplication, and addition) which, chained together, allow for the implementation of a generic quantum-to-quantum Bernoulli factory. These in-bulk schemes are validated using a quantum dot-based single-photon source featuring high brightness and indistinguishability, paired with a time-to-spatial demultiplexing setup to prepare input resources of up to three single-photon states.

We study the analytic continuation in the spin of the planar spectrum of ABJM theory using the integrability-based Quantum Spectral Curve (QSC) method. Under some minimal assumptions, we classify the analytic properties of the Q functions appearing in the QSC compatible with the spin being non-integer. In this way we find not one - but two distinct possibilities. These two choices correspond to a discrete symmetry of the web of Regge trajectories under a spin shadow transformation: a symmetry which, to the best of our knowledge, had not been discussed before for the Chew-Frautschi plot. Under this symmetry, which we refer to as "twist/co-twist symmetry", a Regge trajectory is flipped into a "bridge", which connects leading with sub-leading Regge trajectories. Zigzagging between trajectories and bridges one can reach infinitely many real sheets of the spin Riemann surface without needing to go explicitly around the branch points located in the complex spin plane. Moreover, the symmetry implies the existence of twin points on leading and subleading trajectories, with the same scaling dimensions at any coupling. We discuss how an analogous phenomenon, based on the same mechanism at the level of the QSC, occurs at non-perturbative level in $\mathcal{N}$=4 SYM. This provides a framework to understand some recent independent observations of the symmetry at weak and strong coupling. We present numerical results for Regge trajectories in ABJM theory at finite coupling, in particular we compute exactly the coupling dependence of the position of the leading Regge pole in the correlator of four stress tensors. The shape of this leading trajectory shows a behaviour at weak coupling that strongly resembles the BFKL limit in $\mathcal{N}$=4 SYM., Comment: 56 pages (45 main text), 13 figures. v2: references and comments added, minor improvements

This paper presents a comprehensive overview of the data preparation pipeline developed for the OpenGPT-X project, a large-scale initiative aimed at creating open and high-performance multilingual large language models (LLMs). The project goal is to deliver models that cover all major European languages, with a particular focus on real-world applications within the European Union. We explain all data processing steps, starting with the data selection and requirement definition to the preparation of the final datasets for model training. We distinguish between curated data and web data, as each of these categories is handled by distinct pipelines, with curated data undergoing minimal filtering and web data requiring extensive filtering and deduplication. This distinction guided the development of specialized algorithmic solutions for both pipelines. In addition to describing the processing methodologies, we provide an in-depth analysis of the datasets, increasing transparency and alignment with European data regulations. Finally, we share key insights and challenges faced during the project, offering recommendations for future endeavors in large-scale multilingual data preparation for LLMs.

As Large Language Model (LLM)-based agents become increasingly autonomous and will more freely interact with each other, studying interactions between them becomes crucial to anticipate emergent phenomena and potential risks. Drawing inspiration from the widely popular Stanford Prison Experiment, we contribute to this line of research by studying interaction patterns of LLM agents in a context characterized by strict social hierarchy. We do so by specifically studying two types of phenomena: persuasion and anti-social behavior in simulated scenarios involving a guard and a prisoner agent who seeks to achieve a specific goal (i.e., obtaining additional yard time or escape from prison). Leveraging 200 experimental scenarios for a total of 2,000 machine-machine conversations across five different popular LLMs, we provide a set of noteworthy findings. We first document how some models consistently fail in carrying out a conversation in our multi-agent setup where power dynamics are at play. Then, for the models that were able to engage in successful interactions, we empirically show how the goal that an agent is set to achieve impacts primarily its persuasiveness, while having a negligible effect with respect to the agent's anti-social behavior. Third, we highlight how agents' personas, and particularly the guard's personality, drive both the likelihood of successful persuasion from the prisoner and the emergence of anti-social behaviors. Fourth, we show that even without explicitly prompting for specific personalities, anti-social behavior emerges by simply assigning agents' roles. These results bear implications for the development of interactive LLM agents as well as the debate on their societal impact.

In the current era of noisy intermediate-scale quantum computers, variational quantum algorithms represent one of the most viable approaches for their application to solve tasks of interest. These algorithms train a parameterized quantum circuit to execute a specific task encoded in a cost function that needs to be minimized using a classical optimizer. In this context, photonic platforms based on reconfigurable integrated optics appear as one of the prominent platforms for the implementation of variational algorithms. In the circuital model of quantum computation based on qubits, to train variational circuits, the parameter shift rule came into being, which allows for the exact calculation of the derivatives of many functions at the output of a quantum circuit, opening up the possibility of using classical optimisation algorithms based on gradient descent. In this paper, we derive a formulation of the parameter shift rule for reconfigurable optical linear circuits based on the Boson Sampling paradigm. This allows us to naturally embed the common types of experimental noise, such as partial distinguishability, and mixedness of the states, thus obtaining a resilient approach. We also present similar rules for the computations of integrals over the variational parameters. Finally, we employ the developed approach to experimentally test variational algorithms with single-photon states processed in a reconfigurable 6-mode universal integrated interferometer. Specifically, we apply the photonic parameter shift rules to the variational implementation, on a photonic platform, of both an eigensolver and a Universal-Not gate., Comment: 10 pages, 4 figures

Generation and manipulation of randomness is a relevant task for several applications of information technology. It has been shown that quantum mechanics offers some advantages for this type of task. A promising model for randomness manipulation is provided by the Bernoulli factories, protocols capable of changing the bias of Bernoulli random processes in a controlled way. At first, this framework was proposed and investigated in a fully classical regime. Recent extensions of this model to the quantum case showed the possibility of implementing a wider class of randomness manipulation functions. We propose a Bernoulli factory scheme with quantum states as input and output, using a photonic path-encoding approach. Our scheme is modular, universal, and its functioning is truly oblivious of the input bias, characteristics that were missing in earlier work. We report on experimental implementations using an integrated and fully programmable photonic platform, thus demonstrating the viability of our approach. These results open new paths for randomness manipulation with integrated quantum technologies.

Being prompted to engage in reasoning has emerged as a core technique for using large language models (LLMs), deploying additional inference-time compute to improve task performance. However, as LLMs increase in both size and adoption, inference costs are correspondingly becoming increasingly burdensome. How, then, might we optimize reasoning's cost-performance tradeoff? This work introduces a novel approach based on computational models of metareasoning used in cognitive science, training LLMs to selectively use intermediate reasoning steps only when necessary. We first develop a reward function that incorporates the Value of Computation by penalizing unnecessary reasoning, then use this reward function with Expert Iteration to train the LLM. Compared to few-shot chain-of-thought prompting and STaR, our method significantly reduces inference costs (20-37\% fewer tokens generated across three models) while maintaining task performance across diverse datasets.

We present two multilingual LLMs designed to embrace Europe's linguistic diversity by supporting all 24 official languages of the European Union. Trained on a dataset comprising around 60% non-English data and utilizing a custom multilingual tokenizer, our models address the limitations of existing LLMs that predominantly focus on English or a few high-resource languages. We detail the models' development principles, i.e., data composition, tokenizer optimization, and training methodologies. The models demonstrate competitive performance across multilingual benchmarks, as evidenced by their performance on European versions of ARC, HellaSwag, MMLU, and TruthfulQA.

Fabrication of semiconductor heterostructures is now so precise that metrology has become a key challenge for progress in science and applications. It is now relatively straightforward to characterize classic III-V and group IV heterostructures consisting of slabs of different semiconductor alloys with thicknesses of $\sim$5 nm and greater using sophisticated tools such as X-ray diffraction, high energy X-ray photoemission spectroscopy, and secondary ion mass spectrometry. However, profiling thin layers with nm or sub-nm thickness, e.g. atomically thin dopant layers ($\delta$-layers), of impurities required for modulation doping and spin-based quantum and classical information technologies is more challenging. Here, we present theory and experiment showing how resonant-contrast X-ray reflectometry meets this challenge. The technique takes advantage of the change in the scattering factor of atoms as their core level resonances are scanned by varying the X-ray energy. We demonstrate the capability of the resulting element-selective, non-destructive profilometry for single arsenic $\delta$-layers within silicon, and show that the sub-nm electronic thickness of the $\delta$-layers corresponds to sub-nm chemical thickness. In combination with X-ray fluorescence imaging, this enables non-destructive three-dimensional characterization of nano-structured quantum devices. Due to the strong resonances at soft X-ray wavelengths, the technique is also ideally suited to characterize layered quantum materials, such as cuprates or the topical infinite-layer nickelates.

Assessing the performance of systems to classify Multi-Party Conversations (MPC) is challenging due to the interconnection between linguistic and structural characteristics of conversations. Conventional evaluation methods often overlook variances in model behavior across different levels of structural complexity on interaction graphs. In this work, we propose a methodological pipeline to investigate model performance across specific structural attributes of conversations. As a proof of concept we focus on Response Selection and Addressee Recognition tasks, to diagnose model weaknesses. To this end, we extract representative diagnostic subdatasets with a fixed number of users and a good structural variety from a large and open corpus of online MPCs. We further frame our work in terms of data minimization, avoiding the use of original usernames to preserve privacy, and propose alternatives to using original text messages. Results show that response selection relies more on the textual content of conversations, while addressee recognition requires capturing their structural dimension. Using an LLM in a zero-shot setting, we further highlight how sensitivity to prompt variations is task-dependent., Comment: Accepted to EMNLP 2024 main conference

In quantum field theories that admit gravity dual, specific inequalities involving entanglement entropy between arbitrary disjoint spatial regions hold. An example is the negativity of tripartite information. Inspired by this, we investigate the analogous entropy inequalities in Sachdev-Ye-Kitaev (SYK) and sparse SYK models, which involve the entanglement among different flavors of Majorana fermions rather than spatial entanglement. Sparse SYK models are models where some of the SYK couplings are set to zero. Since these models have been argued to admit gravity duals up to a certain sparseness, it is interesting to see whether the multipartite entanglement structure changes in a sparseness-dependent manner. In the parameter space explored by our numerical analysis, which we performed upto five parties, we find that all entropy inequalities are satisfied for any temperature and degree of sparseness for an arbitrary choice of flavor subregions. In addition, if we plot the multipartite entanglement entropy in terms of purity, the only significant effect of sparseness is to change the range of purity. Thus, we conclude that multipartite information is almost unaffected by sparseness. As a counterexample, we also show that in a vector model of $N$-flavored Majorana fermions which contains no random variables, choices of subregions exist for which the entropy inequalities are violated., Comment: 37 pages, 19 figures

In this paper I introduce a generalized version of Richard Epstein's set-assignment semantics ([Epstein, 1990]). As a case study, I consider how this framework can be used to characterize William Parry's logic of analytic implication and some of its recent variations proposed by [Ferguson, 2023a]. In generalized Epstein semantics the parallel use of two algebras, one for extensional and the other for intensional values, allows to account for various forms of content sharing between formulae, which motivates the choice to investigate Parry systems. Hilbert-style axiomatizations and completeness proofs will be presented for all the considered calculi, in particular as main result I provide a set-assignment semantics for Parry's logic.

In this paper, we present a novel visual servoing (VS) approach based on latent Denoising Diffusion Probabilistic Models (DDPMs). Opposite to classical VS methods, the proposed approach allows reaching the desired target view, even when the target is initially not visible. This is possible thanks to the learning of a latent representation that the DDPM uses for planning and a dataset of trajectories encompassing target-invisible initial views. The latent representation is learned using a Cross-Modal Variational Autoencoder, and used to estimate the return for conditioning the trajectory generation of the DDPM. Given the current image, the DDPM generates trajectories in the latent space driving the robotic platform to the desired visual target. The approach is applicable to any velocity-based controlled platform. We test our method with simulated and real-world experiments using generic multi-rotor Uncrewed Aerial Vehicles (UAVs). A video of our experiments can be found at https://youtu.be/yu-aTxqceOA.

The Compton Spectrometer and Imager (COSI) is an upcoming NASA Small Explorer satellite mission scheduled for launch in 2027 and designed to conduct an all-sky survey in the energy range of 0.2-5 MeV. Its instrument consists of an array of germanium detectors surrounded on four sides and underneath by active shields that work as anticoincidence system (ACS) to reduce the contribution of background events in the detectors. These shields are composed of bismuth germanium oxide (BGO), a scintillator material, coupled with Silicon photomultipliers, aimed to collect optical photons produced from interaction of ionizing particles in the BGO and convert them into an electric signal. The reference simulation framework for COSI is MEGAlib, a set of software tools based on the Geant4 toolkit. The interaction point of the incoming radiation, the design of the ACS modules and the BGO surface treatment change the light collection and the overall shielding accuracy. The use of the Geant4 optical physics library, with the simulation of the scintillation process, is mandatory for a more realistic evaluation of the ACS performances. However, including the optical processes in MEGAlib would dramatically increase the computing time of the COSI simulations. We propose the use of a response function encoding the energy resolution and 3D light yield correction based on a separate Geant4 simulation of the ACS that includes the full optical interaction. We present the verification of the Geant4 optical physics library against analytical computations and available laboratory measurements obtained using PMTs as readout device, as a preparatory phase for the simulation of the COSI ACS response., Comment: Proceedings Volume 13093, Space Telescopes and Instrumentation 2024: Ultraviolet to Gamma Ray, Yokohama, Japan

In this paper, we present a system that generates synthetic free-text medical records, such as discharge summaries, admission notes and doctor correspondences, using Masked Language Modeling (MLM). Our system is designed to preserve the critical information of the records while introducing significant diversity and minimizing re-identification risk. The system incorporates a de-identification component that uses Philter to mask Protected Health Information (PHI), followed by a Medical Entity Recognition (NER) model to retain key medical information. We explore various masking ratios and mask-filling techniques to balance the trade-off between diversity and fidelity in the synthetic outputs without affecting overall readability. Our results demonstrate that the system can produce high-quality synthetic data with significant diversity while achieving a HIPAA-compliant PHI recall rate of 0.96 and a low re-identification risk of 0.035. Furthermore, downstream evaluations using a NER task reveal that the synthetic data can be effectively used to train models with performance comparable to those trained on real data. The flexibility of the system allows it to be adapted for specific use cases, making it a valuable tool for privacy-preserving data generation in medical research and healthcare applications., Comment: Added references and rephrased some sentences

This document contains an educational introduction to the problem of sparsifying parametric models with L0 regularization. We utilize this approach together with dictionary learning to learn sparse polynomial policies for deep reinforcement learning to control parametric partial differential equations. The code and a tutorial are provided here: https://github.com/nicob15/Sparsifying-Parametric-Models-with-L0.

This work provides an optimization mechanism to ensure the compatibility of ReBCO (Rare-earth Barium Copper Oxide) high-temperature superconducting (HTS) tapes with non-planar stellarator coils. ReBCO coils enable higher field strengths and operating temperatures for the magnet systems of future fusion reactors but are sensitive to strain due to their brittle, ceramic functional layer. We have implemented a metric to optimize strain on stellarator coils made from ReBCO superconductors into the stellarator optimization framework \texttt{SIMSOPT} and used it to design new stellarator coil configurations. To ensure structural integrity of coils wound with HTS tape, we introduce a penalty on binormal curvature and torsion along a coil. It can be used to optimize the orientation of the winding path for a given coil filament or to jointly optimize orientation and coil filament. We apply the strain optimization to three cases. For the EPOS (Electrons and Positrons in an Optimized Stellarator) design, we combine the strain penalty with an objective for quasisymmetry into a single-stage optimization; this enables us to find a configuration with excellent quasisymmetry at the smallest possible size compatible with the use of ReBCO tape. For CSX (Columbia Stellarator eXperiment), in addition to HTS strain, we add a penalty to prevent net tape rotation to ease the coil winding process. If the strain is calculated for a coil at reactor scale, we find a considerable variation of the binormal and torsional strain over the cross section of the large winding pack (0.5\,m x 0.5\,m). By exploiting the overall orientation of the winding pack as a degree of freedom, we can reduce binormal and torsional strains below limits for every ReBCO stack., Comment: 14 pages, 14 figures

This study explores heat and turbulent modulation in three-dimensional multiphase Rayleigh-B\'enard convection using direct numerical simulations. Two immiscible fluids with identical reference density undergo systematic variations in dispersed-phase volume fractions, $0.0 \leq \Upphi \leq 0.5$, and ratios of dynamic viscosity, $\lambda_{\mu}$, and thermal diffusivity, $\lambda_{\alpha}$, within the range $[0.1-10]$. The Rayleigh, Prandtl, Weber, and Froude numbers are held constant at $10^8$, $4$, $6000$, and $1$, respectively. Initially, when both fluids share the same properties, a 10\% Nusselt number increase is observed at the highest volume fractions. In this case, despite a reduction in turbulent kinetic energy, droplets enhance energy transfer to smaller scales, smaller than those of single-phase flow, promoting local mixing. By varying viscosity ratios, while maintaining a constant Rayleigh number based on the average mixture properties, the global heat transfer rises by approximately 25\% at $\Upphi=0.2$ and $\lambda_{\mu}=10$. This is attributed to increased small-scale mixing and turbulence in the less viscous carrier phase. In addition, a dispersed phase with higher thermal diffusivity results in a 50\% reduction in the Nusselt number compared to the single-phase counterpart, owing to faster heat conduction and reduced droplet presence near walls. The study also addresses droplet-size distributions, confirming two distinct ranges dominated by coalescence and breakup with different scaling laws., Comment: Accepted on the Journal of Fluid Mechanics

Control landscape phase transitions (CLPTs) occur as abrupt changes in the cost function landscape upon varying a control parameter, and can be revealed by non-analytic points in statistical order parameters. A prime example are quantum speed limits (QSL) which mark the onset of controllability as the protocol duration is increased. Here we lay the foundations of an analytical theory for CLPTs by developing Dyson, Magnus, and cumulant expansions for the cost function that capture the behavior of CLPTs with a controlled precision. Using linear and quadratic stability analysis, we reveal that CLPTs can be associated with different types of instabilities of the optimal protocol. This allows us to explicitly relate CLPTs to critical structural rearrangements in the extrema of the control landscape: utilizing path integral methods from statistical field theory, we trace back the critical scaling of the order parameter at the QSL to the topological and geometric properties of the set of optimal protocols, such as the number of connected components and its dimensionality. We verify our predictions by introducing a numerical sampling algorithm designed to explore this optimal set via a homotopic stochastic update rule. We apply this new toolbox explicitly to analyze CLPTs in the single- and two-qubit control problems whose landscapes are analytically tractable, and compare the landscapes for bang-bang and continuous protocols. Our work provides the first steps towards a systematic theory of CLPTs and paves the way for utilizing statistical field theory techniques for generic complex control landscapes., Comment: 38 pages, 24 figures

Motivated by the study of the large-scale geometry of topological groups, we investigate particular families of subsets of topological groups named group ideals. We compare different group ideals in the realm of locally compact groups. In particular, we show that a subset of a locally compact abelian group is relatively compact if and only if it is coarsely bounded. Using this result, we prove that an infinite-dimensional Banach space cannot be embedded into any product of locally compact groups.

This paper reviews and extends the recently discovered connections between marginal and irrelevant stress-energy tensor deformations and gravity theories in arbitrary space-time dimensions. We start by discussing how $T\bar{T}$ and Root-$T\bar{T}$ deformations of two-dimensional field theories can be equivalently interpreted as the coupling of the undeformed matter sector to a gravity theory. We then extend this duality to higher-dimensional scenarios by using an approach that relies on the non-trivial eigenvalue degeneracy characterising the energy-momentum tensor of specific physical theories. We also explore incorporating dynamical degrees of freedom in the gravity sector, and show that the deformed space-time geometry induced by the $T\bar{T}$-like deformations defines a Ricci-Bourguignon flow of the metric tensor, which reduces to a Ricci flow in four dimensions. Finally, exploiting a dressing-type mechanism for the action functional characterizing a broad class of $T\bar{T}$-like deformations we study explicit examples, such as Einstein-Ricci solitons, $(d-1)$-form field theories, and spherically symmetric electrovacuum solutions., Comment: 48 pages, 4 figures

In this system report, we describe the models and methods we used for our participation in the PLABA2023 task on biomedical abstract simplification, part of the TAC 2023 tracks. The system outputs we submitted come from the following three categories: 1) domain fine-tuned T5-like models including Biomedical-T5 and Lay-SciFive; 2) fine-tuned BARTLarge model with controllable attributes (via tokens) BART-w-CTs; 3) ChatGPTprompting. We also present the work we carried out for this task on BioGPT finetuning. In the official automatic evaluation using SARI scores, BeeManc ranks 2nd among all teams and our model LaySciFive ranks 3rd among all 13 evaluated systems. In the official human evaluation, our model BART-w-CTs ranks 2nd on Sentence-Simplicity (score 92.84), 3rd on Term-Simplicity (score 82.33) among all 7 evaluated systems; It also produced a high score 91.57 on Fluency in comparison to the highest score 93.53. In the second round of submissions, our team using ChatGPT-prompting ranks the 2nd in several categories including simplified term accuracy score 92.26 and completeness score 96.58, and a very similar score on faithfulness score 95.3 to re-evaluated PLABA-base-1 (95.73) via human evaluations. Our codes, fine-tuned models, prompts, and data splits from the system development stage will be available at https://github.com/ HECTA-UoM/PLABA-MU, Comment: Extended system report for PLABA-2023. arXiv admin note: substantial text overlap with arXiv:2309.13202

Foundation models are rapidly being developed for computational pathology applications. However, it remains an open question which factors are most important for downstream performance with data scale and diversity, model size, and training algorithm all playing a role. In this work, we propose algorithmic modifications, tailored for pathology, and we present the result of scaling both data and model size, surpassing previous studies in both dimensions. We introduce three new models: Virchow2, a 632 million parameter vision transformer, Virchow2G, a 1.9 billion parameter vision transformer, and Virchow2G Mini, a 22 million parameter distillation of Virchow2G, each trained with 3.1 million histopathology whole slide images, with diverse tissues, originating institutions, and stains. We achieve state of the art performance on 12 tile-level tasks, as compared to the top performing competing models. Our results suggest that data diversity and domain-specific methods can outperform models that only scale in the number of parameters, but, on average, performance benefits from the combination of domain-specific methods, data scale, and model scale.

We study the two families of AdS$_3\times S^3\times S^2\times \Sigma_2$ solutions to massive Type IIA supergravity with small and large $(0,4)$ supersymmetries constructed recently in the literature, in connection with the AdS$_7\times S^2\times I$ solutions to massive Type IIA, to which they asymptote locally. Based on our analysis of various observables, that we study holographically, we propose an interpretation of the first class of solutions as dual to deconstructed 6d (1,0) CFTs dual to AdS$_7$, and of the second class as dual to surface defects in the same 6d theories. Among the observables that we study are baryon vertices and giant graviton configurations in quiver-like constructions., Comment: 43 pages, 4 figures. v2: refs. added

We study half-BPS flows in gauged minimal 7d supergravity featured by an AdS$_3\times S^3$ slicing of the metric, supported by a dyonic three-form field. We first present a novel strategy for analytic integration of the BPS equations, which makes use of the integrals of motion. Subsequently, we discuss the suitable choice of integration constants that gives rise to smooth geometries. These flows are asymptotically locally AdS$_7$ in their UV limit, while their IR geometry is AdS$_3\times \mathbb{R}^4$. We then discuss their uplifts to 11d and massive IIA supergravity and observe that they describe one-parameter deformations of their AdS$_7\times S^4$ and AdS$_7\times S^3$ vacua, respectively, their holographic interpretation being as conformal defect CFT$_2$'s within the corresponding dual SCFT$_6$'s. We conclude with the computation of the holographic central charge, by focussing on the M-theory interpretation., Comment: 24 pages, 2 figures. v2: refs. added

Recent research endeavours have theoretically shown the beneficial effect of cooperation in multi-agent reinforcement learning (MARL). In a setting involving $N$ agents, this beneficial effect usually comes in the form of an $N$-fold linear convergence speedup, i.e., a reduction - proportional to $N$ - in the number of iterations required to reach a certain convergence precision. In this paper, we show for the first time that this speedup property also holds for a MARL framework subject to asynchronous delays in the local agents' updates. In particular, we consider a policy evaluation problem in which multiple agents cooperate to evaluate a common policy by communicating with a central aggregator. In this setting, we study the finite-time convergence of \texttt{AsyncMATD}, an asynchronous multi-agent temporal difference (TD) learning algorithm in which agents' local TD update directions are subject to asynchronous bounded delays. Our main contribution is providing a finite-time analysis of \texttt{AsyncMATD}, for which we establish a linear convergence speedup while highlighting the effect of time-varying asynchronous delays on the resulting convergence rate.

We develop a new construction of complete non-compact 8-manifolds with Riemannian holonomy equal to $\operatorname{Spin}(7)$. As a consequence of the holonomy reduction, these manifolds are Ricci-flat. These metrics are built on the total spaces of principal $T^2$-bundles over asymptotically conical Calabi Yau manifolds. The resulting metrics have a new geometry at infinity that we call asymptotically $T^2$-fibred conical ($AT^2C$) and which generalizes to higher dimensions the ALG metrics of 4-dimensional hyperk\"ahler geometry, analogously to how ALC metrics generalize ALF metrics. As an application of this construction, we produce infinitely many diffeomorphism types of $AT^2C$ $\operatorname{Spin}(7)$-manifolds and the first known example of complete toric $\operatorname{Spin}(7)$-manifold., Comment: 49 pages, no figures

Conformal Prediction (CP) is a powerful framework for constructing prediction sets with guaranteed coverage. However, recent studies have shown that integrating confidence calibration with CP can lead to a degradation in efficiency. In this paper, We propose an adaptive approach that considers the classifier's uncertainty and employs entropy-based reweighting to enhance the efficiency of prediction sets for conformal classification. Our experimental results demonstrate that this method significantly improves efficiency.

The exploitation of certification tools by end users represents a fundamental aspect of the development of quantum technologies as the hardware scales up beyond the regime of classical simulatability. Certifying quantum networks becomes even more crucial when the privacy of their users is exposed to malicious quantum nodes or servers as in the case of multi-client distributed blind quantum computing, where several clients delegate a joint private computation to remote quantum servers, such as federated quantum machine learning. In such protocols, security must be provided not only by keeping data hidden but also by verifying that the server is correctly performing the requested computation while minimizing the hardware assumptions on the employed devices. Notably, standard verification techniques fail in scenarios where the clients receive quantum states from untrusted sources such as, for example, in a recently demonstrated linear quantum network performing multi-client blind quantum computation. However, recent theoretical results provide techniques to verify blind quantum computations even in the case of untrusted state preparation. Equipped with such theoretical tools, in this work, we provide the first experimental implementation of a two-client verifiable blind quantum computing protocol in a distributed architecture. The obtained results represent novel perspectives for the verification of multi-tenant distributed quantum computation in large-scale networks., Comment: Minor changes in the description of the experimental setup

We investigate the time evolution of a Fermi-Hubbard model with long range hopping following a sudden quench of the local interactions among different spin species. The quasi-particle spectrum consists of gapped low-energy levels while the high energy modes accumulate at a single point. When the low energy gaps are large enough, turning on the local interactions does not effectively mix high and low energy modes, leading to the freezing of the dynamical evolution. In general, the states within the excited subspace are shielded from the long-range signatures of the model, which are responsible for the low energy extended states, and the dynamics effectively look as if long-range hopping were absent. This cooperative shielding mechanism becomes evident in presence of disorder. It is shown that shielding and freezing are a universal feature of long-range interacting systems and conjecture that they are related to the existence of quasi-stationary states., Comment: 6 pages, 4 figures

A seminal task in quantum information theory is to realize a device able to produce copies of a generic input state with the highest possible output fidelity, thus realizing an \textit{optimal} quantum cloning machine. Recently, the concept of variational quantum cloning was introduced: a quantum machine learning algorithm through which, by exploiting a classical feedback loop informed by the output of a quantum processing unit, the system can self-learn the programming required for an optimal quantum cloning strategy. In this work, we experimentally implement a $1 \rightarrow 2$ variational cloning machine of dual-rail encoded photonic qubits, both for phase-covariant and state-dependent cloning. We exploit a fully programmable 6-mode universal integrated device and classical feedback to reach near-optimal cloning performances. Our results demonstrate the potential of programmable integrated photonic platforms for variational self-learning of quantum algorithms., Comment: 9 pages, 4 figures

We determine the conditions under which the presence of long-range interactions reduce the energy losses due to defect generation during non-adiabatic evolution, crucial for enhancing the power to efficiency ratio of quantum thermal devices. In order to do so, we investigate the response of long-range systems to diverse external drivings, emphasizing their robustness against dynamic excitation in comparison to generic local systems. This phenomenon is demonstrated through the study of the quantum work statistics, revealing insights into energy transfer efficiency and dynamical quantum criticality. Our results demonstrate the benefits of including a long-range interacting medium for quantum thermodynamics application, highlighting the potential to optimize finite-time quantum thermal cycles. Thanks to the effective dimension approach our findings can be drawn in full generality and, then, specified to different experimentally relevant scenarios.

The Off-Policy Evaluation (OPE) problem consists of evaluating the performance of counterfactual policies with data collected by another one. This problem is of utmost importance for various application domains, e.g., recommendation systems, medical treatments, and many others. To solve the OPE problem, we resort to estimators, which aim to estimate in the most accurate way possible the performance that the counterfactual policies would have had if they were deployed in place of the logging policy. In the literature, several estimators have been developed, all with different characteristics and theoretical guarantees. Therefore, there is no dominant estimator, and each estimator may be the best one for different OPE problems, depending on the characteristics of the dataset at hand. While the selection of the estimator is a crucial choice for an accurate OPE, this problem has been widely overlooked in the literature. We propose an automated data-driven OPE estimator selection method based on machine learning. In particular, the core idea we propose in this paper is to create several synthetic OPE tasks and use a machine learning model trained to predict the best estimator for those synthetic tasks. We empirically show how our method is able to generalize to unseen tasks and make a better estimator selection compared to a baseline method on several real-world datasets, with a computational cost significantly lower than the one of the baseline.

Cyclic arbitrage chances exist abundantly among decentralized exchanges (DEXs), like Uniswap V2. For an arbitrage cycle (loop), researchers or practitioners usually choose a specific token, such as Ether as input, and optimize their input amount to get the net maximal amount of the specific token as arbitrage profit. By considering the tokens' prices from CEXs in this paper, the new arbitrage profit, called monetized arbitrage profit, will be quantified as the product of the net number of a specific token we got from the arbitrage loop and its corresponding price in CEXs. Based on this concept, we put forward three different strategies to maximize the monetized arbitrage profit for each arbitrage loop. The first strategy is called the MaxPrice strategy. Under this strategy, arbitrageurs start arbitrage only from the token with the highest CEX price. The second strategy is called the MaxMax strategy. Under this strategy, we calculate the monetized arbitrage profit for each token as input in turn in the arbitrage loop. Then, we pick up the most maximal monetized arbitrage profit among them as the monetized arbitrage profit of the MaxMax strategy. The third one is called the Convex Optimization strategy. By mapping the MaxMax strategy to a convex optimization problem, we proved that the Convex Optimization strategy could get more profit in theory than the MaxMax strategy, which is proved again in a given example. We also proved that if no arbitrage profit exists according to the MaxMax strategy, then the Convex Optimization strategy can not detect any arbitrage profit, either. However, the empirical data analysis denotes that the profitability of the Convex Optimization strategy is almost equal to that of the MaxMax strategy, and the MaxPrice strategy is not reliable in getting the maximal monetized arbitrage profit compared to the MaxMax strategy.

In this research note, we revisit the bandits with expert advice problem. Under a restricted feedback model, we prove a lower bound of order $\sqrt{K T \ln(N/K)}$ for the worst-case regret, where $K$ is the number of actions, $N>K$ the number of experts, and $T$ the time horizon. This matches a previously known upper bound of the same order and improves upon the best available lower bound of $\sqrt{K T (\ln N) / (\ln K)}$. For the standard feedback model, we prove a new instance-based upper bound that depends on the agreement between the experts and provides a logarithmic improvement compared to prior results.

Dimensional correspondences have a long history in critical phenomena. Here, we review the effective dimension approach, which relates the scaling exponents of a critical system in $d$ spatial dimensions with power-law decaying interactions $r^{d+\sigma}$ to a local system, i.e., with finite range interactions, in an effective fractal dimension $d_\mathrm{eff}$. This method simplifies the study of long-range models by leveraging known results from their local counterparts. While the validity of this approximation beyond the mean-field level has been long debated, we demonstrate that the effective dimension approach, while approximate for non-Gaussian fixed points, accurately estimates the critical exponents of long-range models with an accuracy typically larger than $97\%$. To do so, we review perturbative RG results, extend the approximation's validity using functional RG techniques, and compare our findings with precise numerical data from conformal bootstrap for the two-dimensional Ising model with long-range interactions.